Main

In your final repo, there should be an R markdown file that organizes all computational steps for evaluating your proposed Facial Expression Recognition framework.

This file is currently a template for running evaluation experiments. You should update it according to your codes but following precisely the same structure.

#Test Branch created
if(!require("EBImage")){
 install.packages("BiocManager")
 BiocManager::install("EBImage")
}
if(!require("R.matlab")){
  install.packages("R.matlab")
}
if(!require("readxl")){
  install.packages("readxl")
}

if(!require("dplyr")){
  install.packages("dplyr")
}
if(!require("readxl")){
  install.packages("readxl")
}

if(!require("ggplot2")){
  install.packages("ggplot2")
}

if(!require("caret")){
  install.packages("caret")
}

if(!require("glmnet")){
  install.packages("glmnet")
}

if(!require("WeightedROC")){
  install.packages("WeightedROC")
}

if(!require("gbm")){
  install.packages("gbm")
}

if(!require("DMwR")){
  install.packages("DMwR")
}

library(R.matlab)
library(readxl)
library(dplyr)
#library(EBImage)
library(ggplot2)
library(caret)
library(glmnet)
library(WeightedROC)
library(gbm)
library(DMwR)

New libraries

if(!require("randomForest")){
 install.packages("randomForest")
}
if(!require("xgboost")){
 install.packages("xgboost")
}
if(!require("tibble")){
 install.packages("tibble")
}
if(!require("ROSE")){
 install.packages("ROSE")
}
if(!require("ggplot2")){
 install.packages("ggplot2")
}
if(!require("tidyverse")){
 install.packages("tidyverse")
}

if(!require("AUC")){
 install.packages("AUC")
}
if(!require("e1071")){
 install.packages("e1071")
}
if(!require("OpenImageR")){
 install.packages("OpenImageR")
}
if(!require("caTools")){
  install.packages("caTools")
}
library(OpenImageR)
library(AUC)
library(e1071)
library(randomForest)
library(xgboost)
library(tibble)
library(ROSE)
library(ggplot2)
library(tidyverse)
library(AUC)
library(e1071)
library(caTools)

if(!require("prediction")){
  install.packages("prediction")
}
if(!require("pROC")){
  install.packages("pROC")
}
library(prediction)
library(pROC)

Step 0 set work directories

set.seed(2020)
setwd("../doc")
# here replace it with your own path or manually set it in RStudio to where this rmd file is located. 
# use relative path for reproducibility

Provide directories for training images. Training images and Training fiducial points will be in different subfolders.

train_dir <- "../data/train_set/" # This will be modified for different data sets.
train_image_dir <- paste(train_dir, "images/", sep="")
train_pt_dir <- paste(train_dir,  "points/", sep="")
train_label_path <- paste(train_dir, "label.csv", sep="")

Step 1: set up controls for evaluation experiments.

In this chunk, we have a set of controls for the evaluation experiments.

• (T/F) cross-validation on the training set
• (T/F) reweighting the samples for training set
• (number) K, the number of CV folds
• (number) gbm.numtrees, the number of trees to use in GBM baseline
• (T/F) process features for training set
• (T/F) run evaluation on an independent test set
• (T/F) process features for test set
• (T/F) return polynomial features matrix only
• (T/F) add polynomial features to starter code features matrix
• (T/F) run gbm baseline model
• (0/1) alpha, alpha=0 for ridge regression, alpha=1 for lasso regression
• (T/F) train ridge model

run.cv <- FALSE # run cross-validation on the training set
sample.reweight <- FALSE # run sample reweighting in model training
K <- 5  # number of CV folds
gbm.numtrees <- 1000 #number of trees to use in gbm
run.feature.train <- TRUE # process features for training set
run.test <- TRUE # run evaluation on an independent test set
run.feature.test <- TRUE # process features for test set
run.poly.feature <- TRUE # process poly features
run.add.poly.feature <- TRUE # and poly features to dist matrix
run.gbm <- TRUE
needs.balanced <- TRUE # balance data for model fitting
model.selection <- TRUE # perform model selection on svm models
run.balanced.data <- TRUE # Whether or not balance the data
train.random.forest <- F # Train Random Forest Model
tune.random.forest <- F # Tune Random Forest Model
alpha <- 0 # ridge regression
train.ridge <- TRUE # train ridge model

Using cross-validation or independent test set evaluation, we compare the performance of models with different specifications. In this Starter Code, we tune parameter lambda (the amount of shrinkage) for logistic regression with LASSO penalty.

lmbd = c(1e-3, 5e-3, 1e-2, 5e-2, 1e-1)
model_labels = paste("LASSO Penalty with lambda =", lmbd)

The amount of the penalty for ridge regression can be fine-tuned using lambda.

lambda = 10^seq(10, -2, length = 100)

Step 2: import data and train-test split

#train-test split
info <- read.csv(train_label_path)
n <- nrow(info) #get number of rows from csv
n_train <- round(n*(4/5), 0) #use 4/5 amount of data for training
train_idx <- sample(info$Index, n_train, replace = F) #grab indexs used for training
test_idx <- setdiff(info$Index, train_idx) # get indexs not used for training

If you choose to extract features from images, such as using Gabor filter, R memory will exhaust all images are read together. The solution is to repeat reading a smaller batch(e.g 100) and process them.

n_files <- length(list.files(train_image_dir,'*jpg'))

# image_list <- list()
# for(i in 1:100){
#    image_list[[i]] <- readImage(paste0(train_image_dir, sprintf("%04d", i), ".jpg"))
# }

Fiducial points are stored in matlab format. In this step, we read them and store them in a list.

#function to read fiducial points
#input: index
#output: matrix of fiducial points corresponding to the index
readMat.matrix <- function(index){
     return(round(readMat(paste0(train_pt_dir, sprintf("%04d", index), ".mat"))[[1]],0))
}

#load fiducial points
fiducial_pt_list <- lapply(1:n_files, readMat.matrix)
save(fiducial_pt_list, file="../output/fiducial_pt_list.RData")

Step 3: construct features and responses

• The follow plots show how pairwise distance between fiducial points can work as feature for facial emotion recognition.

  • In the first column, 78 fiducials points of each emotion are marked in order.
  • In the second column distributions of vertical distance between right pupil(1) and right brow peak(21) are shown in histograms. For example, the distance of an angry face tends to be shorter than that of a surprised face.
  • The third column is the distributions of vertical distances between right mouth corner(50) and the midpoint of the upper lip(52). For example, the distance of an happy face tends to be shorter than that of a sad face.

Figure1

feature.R should be the wrapper for all your feature engineering functions and options. The function feature( ) should have options that correspond to different scenarios for your project and produces an R object that contains features and responses that are required by all the models you are going to evaluate later.

• feature.R
• Input: list of images or fiducial point
• Output: an RData file that contains extracted features and corresponding responses

source("../lib/feature.R")
tm_feature_train <- NA
if(run.feature.train){
  tm_feature_train <- system.time(dat_train<-feature(fiducial_pt_list,train_idx, run.poly.feature, run.add.poly.feature))
  save(dat_train, file="../output/feature_train.RData")
}else{
  load(file="../output/feature_train.RData")
}
tm_feature_test <- NA
if(run.feature.test){
  tm_feature_test <- system.time(dat_test <- feature(fiducial_pt_list, test_idx, run.poly.feature, run.add.poly.feature))
  save(dat_test, file="../output/feature_test.RData")
}else{
  load(file="../output/feature_test.RData")
}
# transfer label column from factor to numeric
dat_train$label <- as.numeric(dat_train$label)-1
dat_test$label <- as.numeric(dat_test$label)-1
#Rebalancing training data-Bootstrap Random Over-Sampling Examples Technique (ROSE) source
if(run.balanced.data){
dat_train_balanced_rose<-ROSE(label~., dat_train,seed=2020)$data
save(dat_train_balanced_rose, file="../output/balanced_data.RData")
}else{
  load(file="../output/balanced_data.RData")
}
table(dat_train_balanced_rose$label)

Step 4: Train a classification model with training features and responses

Call the train model and test model from library.

train.R and test.R should be wrappers for all your model training steps and your classification/prediction steps.

• train.R
  • Input: a data frame containing features and labels and a parameter list.
  • Output:a trained model
• test.R
  • Input: the fitted classification model using training data and processed features from testing images
  • Input: an R object that contains a trained classifier.
  • Output: training model specification
• In this Starter Code, we use logistic regression with LASSO penalty to do classification.

source("../lib/train.R") 
source("../lib/test.R")

Model selection with cross-validation

• Do model selection by choosing among different values of training model parameters.

source("../lib/cross_validation.R")
feature_train = as.matrix(dat_train[, 1:ncol(dat_train)-1])
label_train = as.integer(dat_train$label)
if(run.cv){
  res_cv <- matrix(0, nrow = length(lmbd), ncol = 4)
  for(i in 1:length(lmbd)){
    cat("lambda = ", lmbd[i], "\n")
    res_cv[i,] <- cv.function(features = feature_train, labels = label_train, K, 
                              l = lmbd[i], reweight = sample.reweight)
  save(res_cv, file="../output/res_cv.RData")
  }
}else{
  load("../output/res_cv.RData")
}

Visualize cross-validation results.

# res_cv_rf  <- as.data.frame(res_cv_rf )
# colnames(res_cv_rf ) <- c("mean_error", "sd_error", "mean_AUC", "sd_AUC")
# res_cv_rf$k = as.factor(lmbd)
# 
# if(run.cv){
#   p1 <- res_cv_rf  %>%
#     ggplot(aes(x = as.factor(lmbd), y = mean_error,
#                ymin = mean_error - sd_error, ymax = mean_error + sd_error)) +
#     geom_crossbar() +
#     theme(axis.text.x = element_text(angle = 90, hjust = 1))
# 
#   p2 <- res_cv_rf  %>%
#     ggplot(aes(x = as.factor(lmbd), y = mean_AUC,
#                ymin = mean_AUC - sd_AUC, ymax = mean_AUC + sd_AUC)) +
#     geom_crossbar() +
#     theme(axis.text.x = element_text(angle = 90, hjust = 1))
# 
#   print(p1)
#   print(p2)
# }
# lambda=0.01 is the best

• Choose the “best” parameter value

#par_best <- lmbd[which.min(res_cv_rf$mean_error)] # lmbd[which.max(res_cv$mean_AUC)]

Advanced Models:

Create weight test

label_test <- as.integer(dat_test$label)
weight_test <- rep(NA, length(label_test))
for (i in unique(label_test)){
  weight_test[label_test == i] = 0.5 * length(label_test) / length(label_test[label_test == i])
}
# training weights
weight_train <- rep(NA, length(label_train))
for (v in unique(label_train)){
  weight_train[label_train == v] = 0.5 * length(label_train) / length(label_train[label_train == v])
}

if (run.gbm){
  if (sample.reweight){
    tm_train <- system.time(fit_train <- train_gbm(dat_train, s=0.1, K=K, n=gbm.numtrees,w = weight_train))
  } else {
    tm_train <- system.time(fit_train <- train_gbm(dat_train, s=0.1, K=K, n=gbm.numtrees,w = NULL))
  }
  
  # plot the performance
  best.iter.oob <- gbm.perf(fit_train,method="OOB")  # returns out-of-bag estimated best number of trees
  print(best.iter.oob)
  best.iter.cv <- gbm.perf(fit_train,method="cv")   # returns K-fold cv estimate of best number of trees
  print(best.iter.cv)

} else {
  if (sample.reweight){
    tm_train <- system.time(fit_train <- train(feature_train, label_train, w = weight_train, par_best))
  } else {
    tm_train <- system.time(fit_train <- train(feature_train, label_train, w = NULL, par_best))
  }
}
save(fit_train, file="../output/fit_train.RData")

Step 5: Run test on test images

tm_test = NA
feature_test <- as.matrix(dat_test[, 1:ncol(dat_test)-1])
if(run.test){
  load(file="../output/fit_train.RData")
  if (run.gbm){
    tm_test <- system.time(prob_pred<-test_gbm(fit_train,as.data.frame(feature_test),n=best.iter.cv,pred.type = 'response'))
    
    label_pred <- colnames(prob_pred)[apply(prob_pred, 1, which.max)]
    
  } else {
    tm_test <- system.time({label_pred <- as.integer(test(fit_train, feature_test, pred.type = 'class')); 
                            prob_pred <- test(fit_train, feature_test, pred.type = 'response')})  
  }
  
}

Random Forest:

Tune RF

source("../lib/random_forest.R")
if(tune.random.forest){
time.rf.tune <- system.time(rf.tune <- random_forest_tune(dat_train_balanced_rose))
save(rf.tune, file="../output/rf_tune.RData")
}else(
  load("../output/rf_tune.RData")
)
rf.tune

mtry = 154 is the best.

Find the best ntrees

source("../lib/random_forest.R")

#Train 500
if(tune.random.forest){
time.rf.train <- system.time(random_forest_fit_500 <- random_forest_train_500(dat_train_balanced_rose,mtry = 154))
save(random_forest_fit_500, file = "../output/rf_train_500_trees.RData")
}
#Test 500
random_forest_test_prep=NA
if(tune.random.forest){
 load(file="../output/rf_train_500_trees.RData")
 time.rf.test <- system.time(
   random_forest_test_prep <- random_forest_test(
     model = random_forest_fit_500,testset = dat_test)
               )

random_forest_test_prep <- as.numeric(as.character(random_forest_test_prep))
accu_rf_test <- mean(random_forest_test_prep == dat_test$label)
random_forest_label<-round(random_forest_test_prep)
accu_rf <- sum(weight_test * (random_forest_label == label_test)) / sum(weight_test)
#prob_pred <- lable_pred
tpr.fpr <- WeightedROC(random_forest_test_prep, label_test, weight_test)
auc_rf <- WeightedAUC(tpr.fpr)
cat("The AUC of model after reweighting: RF", "is", auc_rf, ".\n")
cat("The accuracy of model: Random Forest on imbalanced testing data", "is", accu_rf_test*100, "%.\n")
cat("The accuracy of model: Random Forest on balanced testing data", "is", accu_rf*100, "%.\n")
cat("Time for training model Random Forest = ", time.rf.train[1], "s \n")
cat("Time for testing model Random Forest = ",time.rf.test[1], "s \n")
}
# The AUC of model after reweighting: RF is 0.5031999 .
# The accuracy of model: Random Forest on imbalanced testing data is 80.33333 %.
# The accuracy of model: Random Forest on balanced testing data is 50.31999 %.
# Time for training model Random Forest =  20.95 s 
# Time for testing model Random Forest =  0.09 s 

#Train 1000
if(tune.random.forest){
time.rf.train <- system.time(random_forest_fit_1000 <- random_forest_train_1000(dat_train_balanced_rose,mtry = 154))
save(random_forest_fit_1000, file = "../output/rf_train_1000_trees.RData")
}
#Test 1000
random_forest_test_prep=NA
if(tune.random.forest){
 load(file="../output/rf_train_1000_trees.RData")
 time.rf.test <- system.time(
   random_forest_test_prep <- random_forest_test(
     model = random_forest_fit_1000,testset = dat_test)
               )

random_forest_test_prep <- as.numeric(as.character(random_forest_test_prep))
accu_rf_test <- mean(random_forest_test_prep == dat_test$label)
random_forest_label<-round(random_forest_test_prep)
accu_rf <- sum(weight_test * (random_forest_label == label_test)) / sum(weight_test)
#prob_pred <- lable_pred
tpr.fpr <- WeightedROC(random_forest_test_prep, label_test, weight_test)
auc_rf <- WeightedAUC(tpr.fpr)
cat("The AUC of model after reweighting: RF", "is", auc_rf, ".\n")
cat("The accuracy of model: Random Forest on imbalanced testing data", "is", accu_rf_test*100, "%.\n")
cat("The accuracy of model: Random Forest on balanced testing data", "is", accu_rf*100, "%.\n")
cat("Time for training model Random Forest = ", time.rf.train[1], "s \n")
cat("Time for testing model Random Forest = ",time.rf.test[1], "s \n")
}

#Train 1500
if(tune.random.forest){
time.rf.train <- system.time(random_forest_fit_1500 <- random_forest_train_1500(dat_train_balanced_rose,mtry = 154))
save(random_forest_fit_1500, file = "../output/rf_train_1500_trees.RData")
}
#Test 1500
random_forest_test_prep=NA
if(tune.random.forest){
 load(file="../output/rf_train_1500_trees.RData")
 time.rf.test <- system.time(
   random_forest_test_prep <- random_forest_test(
     model = random_forest_fit_1500,testset = dat_test)
               )

random_forest_test_prep <- as.numeric(as.character(random_forest_test_prep))
accu_rf_test <- mean(random_forest_test_prep == dat_test$label)
random_forest_label<-round(random_forest_test_prep)
accu_rf <- sum(weight_test * (random_forest_label == label_test)) / sum(weight_test)
#prob_pred <- lable_pred
tpr.fpr <- WeightedROC(random_forest_test_prep, label_test, weight_test)
auc_rf <- WeightedAUC(tpr.fpr)
cat("The AUC of model after reweighting: RF", "is", auc_rf, ".\n")
cat("The accuracy of model: Random Forest on imbalanced testing data", "is", accu_rf_test*100, "%.\n")
cat("The accuracy of model: Random Forest on balanced testing data", "is", accu_rf*100, "%.\n")
cat("Time for training model Random Forest = ", time.rf.train[1], "s \n")
cat("Time for testing model Random Forest = ",time.rf.test[1], "s \n")
}

#Train 2000
if(tune.random.forest){
time.rf.train <- system.time(random_forest_fit_2000 <- random_forest_train_2000(dat_train_balanced_rose,mtry = 154))
save(random_forest_fit_2000, file = "../output/rf_train_2000_trees.RData")
}
#Test 2000
random_forest_test_prep=NA
if(tune.random.forest){
 load(file="../output/rf_train_2000_trees.RData")
 time.rf.test <- system.time(
   random_forest_test_prep <- random_forest_test(
     model = random_forest_fit_2000,testset = dat_test)
               )

random_forest_test_prep <- as.numeric(as.character(random_forest_test_prep))
accu_rf_test <- mean(random_forest_test_prep == dat_test$label)
random_forest_label<-round(random_forest_test_prep)
accu_rf <- sum(weight_test * (random_forest_label == label_test)) / sum(weight_test)
#prob_pred <- lable_pred
tpr.fpr <- WeightedROC(random_forest_test_prep, label_test, weight_test)
auc_rf <- WeightedAUC(tpr.fpr)
cat("The AUC of model after reweighting: RF", "is", auc_rf, ".\n")
cat("The accuracy of model: Random Forest on imbalanced testing data", "is", accu_rf_test*100, "%.\n")
cat("The accuracy of model: Random Forest on balanced testing data", "is", accu_rf*100, "%.\n")
cat("Time for training model Random Forest = ", time.rf.train[1], "s \n")
cat("Time for testing model Random Forest = ",time.rf.test[1], "s \n")
}
#Train 2500
if(tune.random.forest){
time.rf.train <- system.time(random_forest_fit_2500 <- random_forest_train_2500(dat_train_balanced_rose,mtry = 154))
save(random_forest_fit_2500, file = "../output/rf_train_2500_trees.RData")
}
#Test 2500
random_forest_test_prep=NA
if(tune.random.forest){
 load(file="../output/rf_train_2500_trees.RData")
 time.rf.test <- system.time(
   random_forest_test_prep <- random_forest_test(
     model = random_forest_fit_2500,testset = dat_test)
               )

random_forest_test_prep <- as.numeric(as.character(random_forest_test_prep))
accu_rf_test <- mean(random_forest_test_prep == dat_test$label)
random_forest_label<-round(random_forest_test_prep)
accu_rf <- sum(weight_test * (random_forest_label == label_test)) / sum(weight_test)
#prob_pred <- lable_pred
tpr.fpr <- WeightedROC(random_forest_test_prep, label_test, weight_test)
auc_rf <- WeightedAUC(tpr.fpr)
cat("The AUC of model after reweighting: RF", "is", auc_rf, ".\n")
cat("The accuracy of model: Random Forest on imbalanced testing data", "is", accu_rf_test*100, "%.\n")
cat("The accuracy of model: Random Forest on balanced testing data", "is", accu_rf*100, "%.\n")
cat("Time for training model Random Forest = ", time.rf.train[1], "s \n")
cat("Time for testing model Random Forest = ",time.rf.test[1], "s \n")
}

Testing Result: When trees = 500: The AUC of model after reweighting: RF is 0.5116745 . The accuracy of model: Random Forest on imbalanced testing data is 80.66667 %. The accuracy of model: Random Forest on balanced testing data is 51.16745 %. Time for training model Random Forest = 713.63 s Time for testing model Random Forest = 0.19 s

When trees = 1000 The AUC of model after reweighting: RF is 0.5201491 . The accuracy of model: Random Forest on imbalanced testing data is 81 %. The accuracy of model: Random Forest on balanced testing data is 52.01491 %. Time for training model Random Forest = 1367.94 s Time for testing model Random Forest = 0.28 s

When trees = 1500 The AUC of model after reweighting: RF is 0.5201491 . The accuracy of model: Random Forest on imbalanced testing data is 81 %. The accuracy of model: Random Forest on balanced testing data is 52.01491 %. Time for training model Random Forest = 2077.56 s Time for testing model Random Forest = 0.36 s

When trees = 2000 The AUC of model after reweighting: RF is 0.5201491 . The accuracy of model: Random Forest on imbalanced testing data is 81 %. The accuracy of model: Random Forest on balanced testing data is 52.01491 %. Time for training model Random Forest = 3142.77 s Time for testing model Random Forest = 0.56 s

When trees = 2500 The AUC of model after reweighting: RF is 0.5159118 . The accuracy of model: Random Forest on imbalanced testing data is 80.83333 %. The accuracy of model: Random Forest on balanced testing data is 51.59118 %. Time for training model Random Forest = 3963.67 s Time for testing model Random Forest = 0.62 s

Therefore, we should use trees = 1000.

Train RF with tuning parameters:

source("../lib/random_forest.R")
if(train.random.forest){
  time.rf.train <- system.time(random_forest_fit <- random_forest_train(dat_train_balanced_rose,mtry = 154))
  save(random_forest_fit, file = "../output/random_forest_train.RData")
  save(time.rf.train,file = "../output/random_forest_train_time.RData")
}else{
  load(file = "../output/random_forest_train_time.RData")
  load(file = "../output/random_forest_train.RData")
}

Test RF with tuning parameters

random_forest_test_prep=NA
if(run.test){
 load(file="../output/random_forest_train.RData")
 time.rf.test <- system.time(
   random_forest_test_prep <- random_forest_test(
     model = random_forest_fit,testset = dat_test)
               )
}
## reweight the test data to represent a balanced label distribution
if (run.gbm){
  accu <- mean(dat_test$label == label_pred)
  cat("The accuracy of GBM baseline model is", mean(dat_test$label == label_pred)*100, "%.\n")
  
} else {
  label_test <- as.integer(dat_test$label)
  weight_test <- rep(NA, length(label_test))
  for (v in unique(label_test)){
    weight_test[label_test == v] = 0.5 * length(label_test) / length(label_test[label_test == v])
  }
  
  accu <- sum(weight_test * (label_pred == label_test)) / sum(weight_test)
  tpr.fpr <- WeightedROC(prob_pred, label_test, weight_test)
  auc <- WeightedAUC(tpr.fpr)
  
  
  cat("The accuracy of model:", model_labels[which.min(res_cv$mean_error)], "is", accu*100, "%.\n")
  cat("The AUC of model:", model_labels[which.min(res_cv$mean_error)], "is", auc, ".\n")
}
random_forest_test_prep <- as.numeric(as.character(random_forest_test_prep))
accu_rf_test <- mean(random_forest_test_prep == dat_test$label)

Calculate weightedAUC on testing split

random_forest_label<-round(random_forest_test_prep)
#prob_pred <- lable_pred
tpr.fpr <- WeightedROC(random_forest_test_prep, label_test, weight_test)
auc_rf <- WeightedAUC(tpr.fpr)

Summary of RF

cat("The AUC of model after reweighting: RF", "is", auc_rf, ".\n")
cat("The accuracy of model: Random Forest on testing data", "is", accu_rf_test*100, "%.\n")
cat("Time for training model Random Forest = ", time.rf.train[1], "s \n")
cat("Time for testing model Random Forest = ",time.rf.test[1], "s \n")
#label_test


cat("The accuracy of model:", model_labels[which.min(res_cv$mean_error)], "is", accu*100, "%.\n")
cat("The AUC of model:", model_labels[which.min(res_cv$mean_error)], "is", auc, ".\n")

Summarize Running Time

Prediction performance matters, so does the running times for constructing features and for training the model, especially when the computation resource is limited.

# cat("Time for constructing training features=", tm_feature_train[1], "s \n")
# cat("Time for constructing testing features=", tm_feature_test[1], "s \n")
# cat("Time for training model=", tm_train[1], "s \n") 
# cat("Time for testing model=", tm_test[1], "s \n")

SVM Model

• Balance the Training Data

library(ROSE)
tm_svm_rebalanced_train <- NA
if(needs.balanced){
  tm_svm_rebalanced_train <- system.time(svm_training_data <- ROSE(label ~ ., data = dat_train)$data)
  save(svm_training_data, file="../output/svm_training_data.RData")
} else {
  load(file="../output/svm_training_data.RData")
}

• Model Selection

library(e1071)
tm_svm_default_mod <- NA
tm_svm_linear_cost <-NA
tm_svm_linear_mod <- NA
if(model.selection){
  svm_model_auc <- rep(NA, 2)
  # default model
  if(run.cv){
    tm_svm_default_mod < system.time(svm_default_mod <- svm_default_train(svm_training_data, K))
    save(svm_default_mod, file="../output/svm_default_mod.RData")
  } else {
    load(file="../output/svm_default_mod.RData")
  }
  svm_default_pred <- svm_test(svm_default_mod, svm_training_data)
  #mean(round(svm_default_pred == svm_training_data$label))
  tpr.fpr_default <- WeightedROC(as.numeric(svm_default_pred), svm_training_data$label)
  svm_model_auc[1] <- WeightedAUC(tpr.fpr_default)
  
  
  # linear kernel
  if(run.cv){
    tm_svm_linear_cost < system.time(best.cost <- svm.cv.linear(svm_training_data, K))
    tm_svm_linear_mod <- system.time(svm_linear_mod <- svm_linear_train(svm_training_data, best.cost, K))
    save(svm_linear_mod, file="../output/svm_linear_mod.RData")
  } else {
    load(file="../output/svm_linear_mod.RData")
  }
  svm_linear_pred <- svm_test(svm_linear_mod, svm_training_data)
  #mean(round(svm_linear_pred == svm_training_data$label))
  tpr.fpr_linear <- WeightedROC(as.numeric(svm_linear_pred), svm_training_data$label)
  svm_model_auc[2] <- WeightedAUC(tpr.fpr_linear)
  
  
  # select model with the highest auc
  curr_best_auc <- which.max(svm_model_auc)
  if(curr_best_auc == 1){
    svm_best_mod <- svm_default_mod
    save(svm_best_mod, file="../output/svm_best_mod.RData")
  } else{
    svm_best_mod <- svm_linear_mod
    save(svm_best_mod, file="../output/svm_best_mod.RData")
  }
} else{
  load(file="../output/svm_best_mod.RData")
}

• Evaluation on Testing Data

tm_svm_rebalanced_test <- NA
if(needs.balanced){
  tm_svm_rebalanced_test <- system.time(svm_testing_data <- ROSE(label ~ ., data = dat_test)$data)
  save(svm_testing_data, file="../output/svm_testing_data.RData")
} else {
  load(file="../output/svm_testing_data.RData")
}
tm_svm_test <- system.time(svm_pred <- svm_test(svm_linear_mod, svm_testing_data))

svm_accu = mean(round(svm_pred == svm_testing_data$label))
tpr.fpr <- WeightedROC(as.numeric(svm_pred), svm_testing_data$label)
svm_auc = WeightedAUC(tpr.fpr)

cat("The accuracy of svm model is", svm_accu*100, "%.\n")
cat("The AUC of svm model is", svm_auc, ".\n")

• Summarize Running Time

cat("Time for rebalancing training data =", tm_svm_rebalanced_train[1], "s \n")
cat("Time for rebalancing testing data =", tm_svm_rebalanced_test[1], "s \n")
cat("Time for training model =", tm_svm_linear_mod[1], "s \n")
cat("Time for testing model=", tm_svm_test[1], "s \n")

ridge model

apply constructed ridge model to the training data

tm_ridge_train <- NA
if (train.ridge){
  dat_train_rebalanced <- ROSE(label ~ ., data = dat_train, seed=2021)$data
  tm_ridge_train <- system.time(ridge_cv_model<-ridge_train(train_data=dat_train_rebalanced, alpha=alpha, K=K, lambda=lambda))
  save(ridge_cv_model, file="../output/ridge_cv_model.RData")
  save(tm_ridge_train, file="../output/ridge_train_time.RData")
}else{
  load(file="../output/ridge_cv_model.RData")
  load(file="../output/ridge_train_time.RData")
}

use cross-validation to choose the optimal lambda with smallest MSE

if (run.cv){
  set.seed(2020)
  feature_train = as.matrix(dat_train[, -6007])
  label_train = as.integer(dat_train$label)
  ridge_model = cv.glmnet(x=feature_train, y=label_train, alpha=alpha, nfolds=K, lambda=lambda)
  opt_lambda = ridge_model$lambda.min
  save(opt_lambda, file="../output/ridge_optimal_lambda.RData")
}else{
  load(file="../output/ridge_optimal_lambda.RData")
}

predict testing data with the optimal lambda

tm_ridge_test = NA
if(run.test){
  load("../output/ridge_cv_model.RData")
  feature_test <- as.matrix(dat_test[, -6007])
  tm_ridge_test <- system.time(label_pred<-as.integer(ridge_test(model=ridge_cv_model, features=feature_test, pred.type = 'class')))
  save(tm_ridge_test, file="../output/ridge_test_time.RData")
} else{
  load(file="../output/ridge_test_time.RData")
}

summarize running time

cat("Time for constructing training features=", tm_feature_train[1], "s \n")
cat("Time for constructing testing features=", tm_feature_test[1], "s \n")
cat("Time for training ridge model=", tm_ridge_train[1], "s \n") 
cat("Time for testing ridge model=", tm_ridge_test[1], "s \n")

run evaluation on independent testing data

load("../output/ridge_cv_model.RData")
feature_test <- as.matrix(dat_test[, -6007])
label_pred = as.integer(predict(ridge_cv_model, s=opt_lambda, newx=feature_test, type='class'))
label_test = as.integer(dat_test$label)
compare <- cbind (label_test, label_pred)
ridge_accuracy = mean(apply(compare, 1, min)/apply(compare, 1, max)) 
cat("The accuracy of the ridge model is", ridge_accuracy*100, "%.\n")
ridge_AUC = auc(roc(label_pred,label_test))
cat("The AUC of the ridge model is", ridge_AUC, ".\n")

###Reference - Du, S., Tao, Y., & Martinez, A. M. (2014). Compound facial expressions of emotion. Proceedings of the National Academy of Sciences, 111(15), E1454-E1462.